Sam Altman sympathizes with mathematicians, but says OpenAI’s math onslaught won’t stop

On the Cheeky Pint podcast, Sam Altman acknowledged sympathy for mathematicians overwhelmed by OpenAI's flood of AI-generated results, then said the company won't slow down. He estimated OpenAI has spent on the order of hundreds of millions of dollars on math, including roughly $10 million on its Navier-Stokes work, and confirmed it holds more proofs than it has released.

That is a big statement about where science is heading. A single lab with that kind of budget can now generate results faster than the field can verify them, and it gets to decide what is released, when, and how.

We want to be clear about our starting point: AI that produces real mathematical results is good news. If models can crack problems humans have chased for generations, the world should have those answers. Altman's line that the company should not sit on a major fundamental discovery is a principle we'd defend.

But principles have to survive contact with practice. According to The Verge, mathematicians have repeatedly complained about thin attribution, little context and insufficient verification. This month alone OpenAI released hundreds of results, often with short write-ups researchers describe as hard to follow. That turns the scientific community into an unpaid verification department for a company with a nine-figure math budget.

And the openness argument cuts the other way too. Details of one claimed result involving Yang-Mills theory have not been made public, on the advice of an expert group OpenAI picked, which reportedly contradicts advice from an earlier group it consulted. You can't say the world deserves the knowledge and also decide privately which knowledge it gets. Altman's admission that he does have more unreleased proofs sharpens the tension.

The part that worries me most is money. Researchers fear being priced out of their own fields. If discovery increasingly runs through whoever can spend millions on a single problem, the shape of mathematics starts to follow a corporate roadmap. That is where this leads if it scales: faster answers, fewer people who understand them, and less independence in deciding which questions matter.

The fix is not to stop. It's to release with proofs that can be checked, clear credit to prior human work, and a consistent disclosure policy. OpenAI says it is adopting a new communications approach. We'll judge it by whether mathematicians stop sounding bewildered.

Altman also acknowledged the push on math problems briefly got out of control, describing an internal Slack channel where staff threw the hardest problems they could find at models to test their limits. OpenAI recently announced claimed breakthroughs across mathematics, physics and computer science.